Handy

Handy is an open-source web app and server that provides local speech-to-text transcription. It allows users to transcribe voice input into any active text field using a configurable keyboard shortcut, ensuring that audio processing remains on the local machine rather than being sent to a cloud service. This eliminates the need for internet connectivity during dictation and prevents audio data from leaving the user's hardware.

The software is deployed as a cross-platform desktop application for Windows, macOS, and Linux. It functions as a background utility that captures audio, converts it to text, and pastes the result directly into the focused application. Users can manage settings through a graphical interface or control the application via command-line flags to toggle transcription or cancel operations. It is designed for users who need a simple, privacy-focused tool for accessibility or productivity without a subscription paywall.

Key features

• Local transcription using Whisper and Parakeet V3 models
• Configurable keyboard shortcuts for recording and push-to-talk
• GPU acceleration for Whisper models when hardware is available
• Voice Activity Detection via Silero to filter silence
• Automatic language detection with CPU-optimized models
• Integration with Raycast for managing history and dictionaries
• CLI flags for remote control and startup customization
• Cross-platform support for Windows, macOS, and Linux

Handy is built using the Tauri framework, utilizing a React and TypeScript frontend with a Rust backend for system integration and ML inference. It relies on libraries such as whisper-rs and transcribe-rs for speech recognition, and cpal for audio I/O. On Linux, the application integrates with tools like xdotool, wtype, or dotool to handle text input across different display servers such as X11 and Wayland. The architecture is designed to be extensible, making it suitable for developers who wish to fork the project or customize the transcription pipeline for specific needs.

This tool is positioned as a privacy-focused accessibility utility for users who require offline dictation and a forkable codebase.